Astronomers are obsessed with building larger and larger telescopes. There are two promises that we make with bigger telescopes: that they can see fainter things and that they see more detail. But it's been really hard to follow through on that second promise because of atmospheric distortion.

We have this interesting problem with black holes. What is a black hole? It is a region of space where you have mass that's confined to zero volume, which means that the density is infinitely large, which means we have no way of describing, really, what a black hole is!

The key to proving that there's a black hole is showing that there's a tremendous amount of mass in a very small volume. And you can do that with the motions of stars. The way the star moves around the center of the galaxy is very much like the way the planets orbit the sun.

One of the big mysteries about the black hole at the center of the galaxy is, 'Why don't we see emission from matter falling onto the black hole, or, rather, the black hole eating up its surroundings?'

If you just have a single problem to solve, then fine, go ahead and use a neural network. But if you want to do science and understand how to choose architectures, or how to go to a new problem, you have to understand what different architectures can and cannot do.

Science exists, moreover, only as a journey toward troth. Stifle dissent and you end that journey.

I do think we're at a point in our history where almost all of the big, grand, challenges faced by the human race are those that demand a scientific solution: climate change; access to clean water; over-crowding; plastic waste.

If you had to sum up chemistry in one sentence, it might be this: Atoms need to have full shells of electrons to feel satisfied, and different elements steal, shed, or borrow different numbers of electrons to achieve a full shell.