Earlier quoted context omitted.
A formal spec would be good enough too. But since he's specifying concrete output and graphics as part of the language, it's rather hard to specify binary-compatibility without a reference implementation. This is why there are no interpreted languages without a reference implementation that haven't failed miserably.
> This is why there are no interpreted languages without a reference implementation that haven't failed miserably. Matlab? I'm not exactly sure what you mean by "reference implementation", though.
Far from desirable and this is probably Wolfram's wet dream.
I doubt it will happen though.
This can be considered a commercial success, although it's pretty much a failure for the user, the student, academia, ... everybody else.