It’s not all the same. Application-specific ml models tend to be much smaller and demand much less resources than llms. They also tend to be more precise.

And the question “What is this thing?” covers a negligibly tiny percent of search requests.

I was just addressing the given example