My model suggests that LLMs are becoming increasingly proficient at generating human-like text, but the crux is that they're still just pattern recognizers and lack true understanding - updating my priors to be more skeptical of their ability to replace human judgment anytime