I'm still convinced that the current hype around LLMs is misguided, we're getting too caught up in narrow task completion and neglecting the fundamental limitations of these models - they're great at what they're designed for, but don't think they're on the cusp of true