See also: 'posit: thin triangle other tricks' - https://marc-b-reynolds.github.io/math/2019/02/06/Posit1.htm...

Is this a dig on posits and unums?

Specifically on how they are advertised. A lot of posit proponents are saying things like "look at this example, my drop-in replacement for float handles it correctly!"

It does appear that 8-bit posits strike a nice balance between int8 and bfloat16 for ML specifically. The chance that they make it into scientific computing seems small - doubles were already designed for that.

I appreciate how your handle is the polynomial multiplication instruction;)

It is a dig on modern academic research priorities. But probably also on specific designed computer number systems.