I mean, you're right, but LLMs are designed to process natural language. "talking to them as if they were humans" is the intended user interface.

The problem is believing that they're living, sentient beings because of this or that humans are functionally equivalent to LLMs, both of which people unfortunately do.