I believe that, from the academic standpoint, approach 3) leads to nowhere. ... Finally, utility should be anyways based on some measure. A good measure should satisfy a set of well-defined axioms (e.g., L2 does). I doubt it is possible to extend the semantic definitions without violating that. All in all, approach 3) risks becoming a technique that appeals to heuristic (and ML-based) judgments, but is far from being mathematically provable. - tlooto, The Most Powerful AcademicGPT

Framing Your Concern ... Analysis of Approach (3) vs (1)/(2), including why (3) is attractive but struggles scientifically: optimality, generalizability, expanding semantics, axiomatic utility, and re - tlooto, AI-Powered Assistant for Academic and Research