A stub, building some on this tweet…
We may soon be able to compare Microsoft’s new “Deep Search” in Bing with Perplexity’s Copilot (from May 18th).
I feel like I’ve seen some one-off experimentation in UI (v. RLHF) for disambiguation / clarification in some of the other generative search tools but can’t recall.
One term for the sort of searching supported by Perplexity AI’s Copilot and, perhaps, Bing’s Deep Search, may be in-depth-search (including both scoping and drilling-down with multiple refined queries). But some of the advantages highlighted by the reporting on both are around clarifying questions and disambiguation.
It is worth also thinking about the difference between clarifying questions and disambiguation (some of which is handled in related searches or People also asked(PAA) sections) and the prompting and multiple queries on the back-end.
To do this well likely requires work not only in the user interface (UI) but also in the training (inclusive of fine-tuning), and of course the crawling, parsing, and indexing as well.