Mathematical Justifications

One possible approach could be to abstract away (to delegate) all mathematical operations, calculations, etc. to a Mathematical (Language ?) Model.

If we do it, how do we make sure that the gotten results are correct?

Different math benchmarks evaluate performance of LMs on math bench datasets, scoring results in % score.

Probabilistic nature

(What about probabilistic nature of results? Can this nature be different, such that it is precise?)

If yes, what do we consider precise? Precision?

Possible verifications (?)