SemanticSpace

Two opinion scores. Jev is asked where each phrase falls on each concept, from 0 at the opposite to 100 at a defining example, and the plotted point is that pair. The page says so directly: a score is an opinion reflecting common associations as one model sees them, not a fact or a measurement. It positions itself as the readable alternative to unnamed embedding dimensions.

No, and the page separates the two explicitly. Confidence shows how firmly Jev settled on a score, not whether the score is correct: shark reads 97% confident on dangerous and only 60% on big. Reading that number as correctness is the easy mistake to make here.

Not exactly. The page warns that scores can wobble, so asking the same thing twice can differ slightly and a point may shift by a point or so. That is fine for exploring a concept, but it rules out using positions as stable keys or diffing two runs as though they were measurements.

The middle band usually means not applicable rather than uncertain: it is reserved for things that are neutral, mixed or unrelated to the concept, and Jev works from five reference points per axis that it can land anywhere between. Closeness on the map depends entirely on which two concepts you chose, so swapping an axis moves every point.

You name two concepts and Jev judges where each phrase falls on each, from the opposite at 0 to a defining example at 100. It works from five reference points (the opposite, leans away, neutral or mixed, leans toward, defining example) and can land anywhere between them, so kitten at 1 or shark at 99 on dangerous is a direct judgment, not an embedding coordinate.

An embedding is often a thousand or more numbers a computer worked out for itself, none of which has a name, so you can measure closeness but not read why. Here each point is two direct judgments you named, which is why both axes are readable and any concept you can put into words can become one.

Because closeness reflects similar answers, not similar spelling or appearance. Wolf and shark sit as neighbours because both are judged fairly big and very dangerous, not because they look alike, and swapping either concept moves every point, so the map only means anything relative to the axes you chose.