1 Comment

Theorem provers in geometry are algebraic in nature and use the mathematics of algebraic geometry to prove or disprove a theorem. Readers of this newsletter may be interested that a similar approach (theorem proving) is possible in unsupervised evaluation. Its safety utility is that it has a complete set of postulates and thus makes it possible to prove or disprove evaluations of noisy AI agents that occur in an unsupervised setting (unlabeled data). Check out the Jupyter notebooks explaining this logic of evaluation with the "ntqr" Python package - https://ntqr.readthedocs.org/en/latest

Expand full comment