A so-far weaker version of this likely works between LLMs and humans (p(0.97), though much lower that it’s easily detectable already). Humans build predictive models of people they are talking to and copy patterns subliminally too, a bunch of persuasion works this way.
A so-far weaker version of this likely works between LLMs and humans (p(0.97), though much lower that it’s easily detectable already). Humans build predictive models of people they are talking to and copy patterns subliminally too, a bunch of persuasion works this way.
LLM persuasion is going to get wild.