Support Meta's 17-hour sinusoidal smear? about t-a-i HOT 3 CLOSED

It might be that what we really need to maintain internally is not a Rat (a ratio of BigInts) but some kind of symbolic equation, which changes and allows operations to be applied. And, Segment would cease to be a purely linear thing and return more of those symbolic equation things. We could then use that equation to generate as many digits as necessary at output time - be they nanoseconds, milliseconds, or whatever. It would have to be a finite number of digits, of course.

from t-a-i.

Sure, there are better "precise" computable numbers, such as algebraic numbers represented by a polynomial and a segment that isolates a single root, with polynomial coefficients and points represented as rational numbers. The problem is that their implementation is already quite complex, involving polynomial-valued matrices, and operations involving sin/cos would make it much worse, bringing multivariate polynomial systems and Groebner bases into the picture.

I don't think there is a reason to dive into this for special smearing models, unless somebody comes up with any reasoning behind them. Google's choice of linear smearing seems to be a reliable solution. Not that it really matters on frontend...

from t-a-i.

My feeling of late is that Meta is irrelevant.

from t-a-i.