I find all mainstream attempts lacking. I feel as if the app is optimizing by AB tests across the whole population instead of trying to find a proper descriptor of my learning behavior and influencing the teaching actions according to that.

For example, Duolingo transformed into a tapping app where spaced repetition took a backseat and learners are incentivized to tap the same lesson until the badge tells them they know the lesson. The lessons are skewed towards the easier target->source questions because the population finds source->target too hard.

So the app becomes extremely useless to me but I guess useful for the masses.