What worked for me was starting with some simple one, a bit technical, but not over the top. I think the first paper I really read, and this was a couple months ago, was Facebook's "Gorilla: A Fast, Scalable, In-Memory Time Series Database"[0]

After that, I started to write 1-3 paragraph summaries to myself, so that I could always come back to the main takeaways, because I know I'll forget it after 1 week.

And then, last week I created a Twitter account where I post these opinionated summaries [1]

[0] https://www.vldb.org/pvldb/vol8/p1816-teller.pdf

[1] https://twitter.com/paper_takeaways