Will we get to a point when an LLM is even used for cases where other methods are more suitable but would be more difficult to implement?