If we're talking about raw programming skill, I would say that actually learning these fancy algorithms will pay off more than (random) tinkering. It's the difference between directed and undirected practice. Undirected practice only takes you so far. To truly get good at something one must do deliberate, directed practice consistently.
I've learned many algorithms and implemented most of the ones I've learned, but if I haven't recently coded them repeatedly or studied them rigorously, I likely won't be able to code them up on the spot when asked to. But because I learned them in the past, I would be able to recognize which algorithm (from the ones I've learned) best suits the problem.
If you understand algorithms in general, then I think thats enough. Its sort of like being able to play a single song on the guitar is great, but to have the skill to play any song is even better. On the same token, being able to write any algorithm is a much better skill than being able to write out a single algorithm.
Nailed it.
Another interesting thing I noticed about myself in technical interviews is that I have a lot of trouble doing things that I would have no trouble with if I'm tutoring someone. Perhaps it's the anxiety, and practice will surely help, but see above, I'd rather spend my time tinkering with stuff than preparing for interviews.