What is surprising in it? Do you mean we can mutate it’s output by putting the right words, making our queues more efficient? Or that it twists things and hallucinates when it can’t get the right answer? It’s a dumb machine built on large volumes of text and hard statistics. We are learning to work with this behemoth, but none of these features are magic. Neither are biomachines like you and me, but there’s a huge difference between dumb parroting LLMs and human intelligence. They don’t even understand what’s the logic behind their output.