3 ms·
It's cute math, but doesn't take into account the inexplicable differences between the retention of seemingly similar facts. Some items stick very well, whereas
by kazinator 4y ago
It's cute math, but doesn't take into account the inexplicable differences between the retention of seemingly similar facts. Some items stick very well, whereas others suffer from lapses.
In my Ankid ecks, I have items that have 10+ year intervals. And some, which were introduced at around the same time as those, which have intervals in months, beaten down by lapses.
Anyone who thinks that some simple math leads to infinite recall has not suddenly lapsed on a card whose interval had reached 7 years.
- orangepurple 4y agoI don't recommend allowing Anki items to have an interval of over one year because the crude SM-2 algorithm does not capture the per-item effect of exponential decay in memory. Multiplying interval lengths by per-item ease factors is suboptimal and not the same. Unlike, for instance, ACT-R, which unfortunately has no good publicly available user interface. A good implementation of a neural memory fact model will track strengths of associative memories in addition to the per-item exponential forgetting curve. Keep the Anki intervals capped at one year until something better comes along.
- kazinator 4y agoProblem is it seems you're then throwing under the bus all those items you recall perfectly well that have 6+ year intervals, which don't require yearly review. Arguably, you could just retire those; but then sometimes they lapse, as I noted: how do you decide what to retire?
- orangepurple 4y agoBuild a knowledge graph of associative memories, measure their strength, and prune clusters with excessive memory strength