Do you think humans don't just "churn out text" at a basic level? Do you think humans don't make mistakes or don't confabulate? The confusion is not that we think too lowly of LLMs rather than that we think we are very special as humans. LLMs like GPT-4 have "sparks of AGI" in them, they are not just stochastic parrots. There are several academic papers that prove LLMs uderstand meaning. GPT-4, the full model, not the crippled chat variant for consumers, demonstrated superhuman coding abilities when researchers probed it's abilities. It passed Google and Amazon coding interviews with 100 out of 100 score and finished the 2 hour assignment in 3minutes surpassing all human candidates. And it took it so long (3m) only because a human had to paste in the tasks.