michael_nielsen

I added several hundred Anki cards while doing.

The early cards were mostly very simple things: facts about TDGammon (which used a similar approach to beat Backgammon), very basic facts about how Go works, and reinforcement learning and Monte Carlo Tree search.

Later, the facts got more and more complex.

Of course, I didn't master all the literature around the paper. But I think I made pretty rapid progress coming up to speed.

Now, just to write one article that wouldn't necessarily have been a good use of time. But a nice thing about Anki is that the information is retained. When the AlphaGo Zero and AlphaZero papers came out, they were very easy to read.

Verb form: I talk and think of "Ankifying" a paper or book etc.

It's easy to overdo it, especially initially, and waste time Ankifying useless info. Over time, I've found myself cultivating heuristics for how much to Ankify.

E.g., for many papers the answer is 0 to 5 questions. But for papers I want to understand better it might be 20 to 40. And for deep papers that I want to master it may be hundreds.