Loading slide
Start with why the last bottle worked, because the reason is narrower than it seems.
That wine was new, but everything about it was familiar. Plum at 0.8, apricot at 0.2. The model had never met that mixture, but it had met plum, and it had met apricot, and it had been corrected on both many times. All it had to do was combine what it already knew in a proportion it had not seen.
Here is a different kind of stranger. A plain cabernet from a legendary
The four units report: cabernet 1, viognier 0, plummy 0, apricotty 0.
Now compare that with the fifth bottle of the training set, an ordinary cabernet with no fruit, which cost 14 dollars. Its four numbers are cabernet 1, viognier 0, plummy 0, apricotty 0.
They are identical. As far as this model can tell, these are the same wine.
So it says 14 dollars, and by its own reasoning it is right. For a wine with those four numbers, 14 dollars is exactly what the eight bottles taught.
That is the difference between the two strangers. The plummy cabernet was a new mixture of things the model can sense. The vintage is a wine whose entire value sits in something it cannot sense at all: there is no unit for the year it was made, no unit for who made it, no unit for how few bottles remain. None of that can reach the price, because there is no wire for it to travel down.
The first kind of stranger it handles. The second it cannot even see coming. Nothing about the year, the maker, or the rarity has any way in, because there are four units and none of them is for those things. A machine that could price this bottle would need more wires than this one has.
Loading contents...