Well, primarily one thing very loudly. That these GenAI systems are not actually that complicated. That when you boil them down, you're usually going to get "person inputs a thing, it jumbles the input up a bit (or if an LLM finds the statistically likely word to go next), then spits it out".