You’re still confusing the infinite Shakespeare monkeys with new ideas. Because

You’re still confusing the infinite Shakespeare monkeys with new ideas. Because there is no idea in the LLM. The idea would come from the individual reading the output of the LLM. And there is no way for the LLM to discover an idea because the only context is the previous uses of the word not the world model.
This is why programming works via LLMs
There are a tiny number of design problems.
Programmatic vocabularies and operations are trivial.
Everything else is syntax
The recombinants fail for the same reason compilation fails -closure.
But there is no closure by action in reality under realism naturalism and operationalism.