Floating-point arithmetic is obsolete

https://www.youtube.com/watch?v=N05yYbUZMSQ
Posits are to be used as an imitation of real numbers just like floats but are almost always more accurate in practice thanks to a non-uniform distribution of relative accuracy. Also, they're easier to implement in hardware and have a larger dynamic range (but that's pointless). MIT licensed. It seems to be working better for deep learning and PDE solving. Intel and NVidia are among interested parties. If this guy can be trusted, are floats already dead?

No thanks, I use Peano numbers.

cool

intervals

*NJ Wildberger voice*
so-called real numbers

https://www.johndcook.com/eightbit.html
posits are rationals

>>6
±∞ is not rational

>>7
It signifies an overflow condition.

There exists no domain where infinity is a number.

>>9
https://en.wikipedia.org/wiki/Riemann_sphere

>>4
*NJ Wildberger voice*
so-called Spherical Earth conjecture

1:02:12

Do you expect us to sit through this whole talk? Give the link to their paper! It can be at least seen by reading every 10th line.

https://www.johndcook.com/blog/2018/04/11/anatomy-of-a-posit-number/
https://posithub.org/index
http://www.johngustafson.net/pdfs/BeatingFloatingPoint.pdf
https://en.wikipedia.org/wiki/Unum_(number_format)

https://en.wikipedia.org/wiki/Interval_arithmetic

I totally agree, floating point arithmetic is obsolete. People should use fixed point numbers instead.
Pros:
- they work with just regular integer arithmetic, you do not need to support special instructions
- More accurate than floats

Cons:
- They are not Java .Net Enterprise Bean Factory enough