current LLMs are impressive but ultimately feel like high-tech parrots - still waiting for one that can have a real conversation without regurgitating some variant of its training data