> I know this might be a silly question but what exactly do you do in an academic research position?
Are you a professor? Then you do whatever you want. If you're not a professor, you do whatever the professor wants.
> How do you choose a topic to look at?
The intersection of what interests you, what you can actually execute on, and what you can publish.
> What do you study on this topic and what is the output supposed to look like?
You figure things out and tell a story.
> I'm assuming I can't just bring in a flash drive and say "Yea I did some reading and did some stuff, here is the implementation of my new X for the Linux kernel. It speeds things up by a factor of BIG_NUMBER."
That's a systems paper! You have to write up a bunch of stuff about what your implementation does and why it's cool, why it's correct, why it's safe, perhaps prove those things, and empirically evaluate BIG_NUMBER in a well controlled and designed experiment.
Are you a professor? Then you do whatever you want. If you're not a professor, you do whatever the professor wants.
> How do you choose a topic to look at?
The intersection of what interests you, what you can actually execute on, and what you can publish.
> What do you study on this topic and what is the output supposed to look like?
You figure things out and tell a story.
> I'm assuming I can't just bring in a flash drive and say "Yea I did some reading and did some stuff, here is the implementation of my new X for the Linux kernel. It speeds things up by a factor of BIG_NUMBER."
That's a systems paper! You have to write up a bunch of stuff about what your implementation does and why it's cool, why it's correct, why it's safe, perhaps prove those things, and empirically evaluate BIG_NUMBER in a well controlled and designed experiment.