Comment by baconbrand

From the LLM's "point of view" it is going to do what characters in the training data were most likely to do.

I have a lot of issues with the framing of it having a "point of view" at all. It is not consciously doing anything.