If we try and map embedding cosine similarity to the human understanding of similarity it doesn't quite match. So this app tries to show cosine similarity up front. A transparent look at it rather than trying to kludge something on top. It's a mental model thing.
No one could accuse Building 20 of burying its Services too deep in the Structure. Recabling from office to office, lab to lab, or even wing to wing is largely a matter of do-it-yourself. Rather than a burden, the occupants consider this a benefit. The wide wood stairs in Building 20 show wear in a way that adds to its myth. You feel yourself walking in historic footsteps in pursuit of technical solutions that might be elegant precisely because they are quick and dirty. And that describes the building: elegant because it is quick and dirty.
What does it take to build something so that it’s really easy to make comfortable little modifications in a way that once you’ve made them, they feel integral with the nature and structure of what is already there?
This book first arose out of a passage in Borges, out of the laughter that shattered, as I read the passage, all the familiar landmarks of my thought—our thought that bears the stamp of our age and our geography—breaking up all the ordered surfaces and all the planes with which we are accustomed to tame the wild profusion of existing things, and continuing long afterwards to disturb and threaten with collapse our age-old distinction between the Same and the Other.
animals are divided into: (a) belonging to the Emperor, (b) embalmed, (c) tame, (d) suckling pigs, (e) sirens, (f) fabulous, (g) stray dogs, (h) included in the present classification, (i) frenzied, (j) innumerable, (k) drawn with a very fine camelhair brush, (l) et cetera, (m) having just broken the water pitcher, (n) that from a long way off look like flies
One of the ways to design a simple user interface that can really do things is to take the hardest tasks imaginable and try to create convincing scenarios of how they might be done at all. Then the simplest tasks must be considered in the light of the structures postulated for the most complex.
Embedding similarity works close to how we imagine search should work. But it can be confusing because things like negation don't work intuitively, and searching at different concept levels can be tough too - "a bird" will probably be closer to "a dog" than "an animal". Makes sense at a certain level. But might not be what you expect. Cosine is trying to show similarity up front and all over. To build that intuition.