> honestly I would not consider the entire context window to be part of a single event, since the model generates tokens one at a time, separately. OK but trad

> honestly I would not consider the entire context window to be part of a single event, since the model generates tokens one at a time, separately. OK but traditional small language model Markov chains have context windows too, and they also generate tokens one at a time, separately. Maybe you are arguing that there is a qualitative difference between an eight token context window and an eight thousand token context window. I guess that would be fair enough.