Comment by thomashop

Funnily, even two different LLMs, when put in conversation with each other, can end up completing each other's sentence. I guess it has something to do with the sequence prediction training objective.