They could also recycle responses. For the majority of search queries, I doubt the answer has changed much in the last week or even month.
It’s actually not that trivial to store & query those responses.
Probably could be done cheaper than running inference on a low cost LLM like Google is doing, but it could be not worth the engineering challenge.
It’s actually not that trivial to store & query those responses. Probably could be done cheaper than running inference on a low cost LLM like Google is doing, but it could be not worth the engineering challenge.