Also LLMs are not capable of actual synthesis … they can only mimic what synthesis sounds like.
That can be useful in some contexts but only a fraction of the use cases the LLM “industry” is pushing on us.
When we ask an LLM to do the process for us, we gain none of the benefit of having done it. Instead of a rough draft for us to polish and reveal the diamond in the rough, what we get is a polished final draft of the statistically most banal and average interpretation of our data.