Solve real-world problems offline using local models. Set yourself constraints: no internet, standard hardware, not a lot of RAM. Research how productive terminal applications have been in the history of computing or why grayscale or even 1-bit user interfaces worked.
That's what I'm doing.