This also seems to be great for low-bandwidth mobile connections. The advantage over reader mode here is that only the plain text is sent and not all the images, large styles and scripts…

The point of these lectures is mostly not to teach how to work with Turing machines, it is to understand the theoretical limits of computers. The Turing machine is just a simple to describe and well-studied tool used to explore that.

For example, are there things there that cannot be computed on a computer, no matter for how long it computes? What about if the computer is able to make guesses along the way, can it compute more? Because of this comic, no — it would only be a lot faster.

Arguably, many programmers can do their job even without knowing any of that. But it certainly helps with seeing the big picture.

So far I have had good experience with kopia. But it is definitly less battle-tested than the other alternatives and I do not use it for too critical stuff yet.

As others said, Synapse can sometimes be very resource-hungry. It might be worth giving a try to Conduit and Dendrite, which are alternative Matrix server implementations and especially Conduit seems to be focused to by lightweight. Although I do not have any personal experience with them and it seems that they are most likely a lot less mature than Synapse at the moment.

Also see pairdrop, it is a snapdrop fork that allows connecting devices on different networks using a numeric code and has other improvements.