As a father trying to get my kindergartner to read and also someone working in ML, it's amazing to me how this mirrors my life experience and ml concepts.

When I was a kid, there was a big effort to experiment on our grade using a concept called whole language, as compared to phonics for reading. I am a whole language person and I've learned to read and retain pretty quickly.

Anyways, this totally mirrors the concept of tokenization. Phonics vs whole languge is suspiciously similar to letter, word, and subword tokenization. One wonders if we as human do a proxy for BPE in our brains when we learn to read.

To this day I'm an absolutely shitty speller.