Honestly, I’ve come up with something like twenty conceptual buckets for future essays, and this one just happened to be the one that collected a dozen examples first; most of the essays in this sequence (if it gets written) will be of similar form. I think raw training data beats pared-down explanation at least half of the time.
(I’m reminded of Ben Pace saying that he never understood bucket errors until my essay that was primarily five examples in a row.)
Honestly, I’ve come up with something like twenty conceptual buckets for future essays, and this one just happened to be the one that collected a dozen examples first; most of the essays in this sequence (if it gets written) will be of similar form. I think raw training data beats pared-down explanation at least half of the time.
(I’m reminded of Ben Pace saying that he never understood bucket errors until my essay that was primarily five examples in a row.)
Could your “twenty conceptual buckets” be specific instances of an even larger bucket?
Yes, “Defense Against The Dark Arts” =P