Comment by famouswaffles

> I fail to see how you can make it work when doing LLM inference which is significantly costlier than web search.

The median LLM query isn't significantly costlier than web search.