It’s also pretty well known in math and CS/SE communities. Heck, the mathematical research into the power of language, and in particular the ability to embed m

It’s also pretty well known in math and CS/SE communities. Heck, the mathematical research into the power of language, and in particular the ability to embed models of reasoning into equations to allow you to formalize your meta-reasoning, is what led to the birth of electric computers. So I find it shallow skepticism to suggest there’s not evidence that it would be useful to AI, because literally the most cursory glance would show mounds of it. The real challenge is and always has been how to combine or embed symbolic reasoning capabilities efficiently within a fuzzier approximator, allowing for “getting close” in a fuzzy way then fine tuning it using more advanced rules. That mimics more closely how we think, where we broadly intuit a few potential answers, then use higher order thinking to sift through them.