I should stress that like any software developer (or any power user, probably), my queries tend to be very specific. I'm looking for particular information (which may or may not exist) that takes more than a couple of words to describe. It is with these queries that all too often Google simply decides that I meant something else and replaces one or two words with what it has decided are synonyms. Except they aren't.
Unfortunately I have not documented all these queries (I should have). But to give some examples, it would not be the first time that Google misinterprets a specific query about "sqlite" and only returns results related to SQL Server. Or it decides that a "scapegoat tree" is equivalent to an "AVL tree". They aren't synonyms. The results are actually worse than irrelevant.
It's like typing a query such as "honda brake problems" and getting results that are about "toyota brake problems". Yes, both are Japanese car manufacturers, but they are really not equivalent.
Of course correcting typing errors and searching for synonyms is useful. But I'd like to have control over that behaviour, and most of all: I want to understand what the search engine does. The best interfaces are those where a user can develop a mental model that helps them interact with it. It seems I'm no longer able to create a mental model for how Google works.
So who do you think should invest the extra effort into performing searches (aka adding quotations.. or "taking them off"): the hackers with specific, but edge-case search requirements, or the 99% of casual users who just want to learn about "how do helycopters fly"
P.S. I tried both the sqlite and the scapegoat tree searches without the quotes and Google seemed to "know" what I meant... but if you encountered that problem perhaps it's an algorithm that unnecessarily "kicks in" depending on some other circumstances of the search. I would definitely classify this as a problem, but one that should again be easily solved with the use of quotation marks
By the way, the scapegoat tree search query was "scapegoat tree inorder append". Try that and observe 90% AVL tree results on the first page.