Nice little example. Could add a "stop" token at the end of the fruit example, so when the little model runs it can call its own stop. To generalize to ADHD,

Nice little example. Could add a "stop" token at the end of the fruit example, so when the little model runs it can call its own stop. To generalize to ADHD, create another model that can quietly "Listen" to a speaker (token generator) until its Markov chain predicts a "Stop", and starts its own generating model running, regardless of whether the speaker was done. Then two instances can have time efficient interrupty conversations with each other.