But then, I feel like it doesn't really scale super well, and doesn't always handle vagueness. For example, even of HN I doubt there will be very many quality/useful answers, simply because there's a certain kind of popularity required to get that here. Looking at something like Quora, there's too much room for non-answers and what amounts to spam. Then, there's the difficulty of being specific enough that people can answer but also do so in a timely manner (the speed of a search engine to provide simply a "close enough" result is part of its utility).
Like the bibliography suggestion, although I said practicality doesn't matter, its an example of how interesting stuff can be discovered (without asking anyone) but requires a huge amount of time (after the initial "look up") and cognitive effort (but that's a kind of curve that tapers off once one is familiar with a particular area of research).
alternatively prine method: look up authors from a set of starter bibliographs & keep authors with higher sitations/lower citations.
can use google to find number of times web resource linked/cited & use the count as a pruning method.
Would be nice for the output to be a kind of collage of abstracts. Maybe convert them to audio using one of these magic AI voice synthesis things and spatially position them in a virtual lecture hall - walk past a door, hear the abstract, walk inside to get the full paper. There could be hallways of doors that are grouped together by some kind of metric.