Comment by lo_zamoyski

That's not quite accurate. Formal languages (which have an old pedigree) can be useful for clarification and inference, but they can also obfuscate the truth, and what's more, subvert it. Every logical formalism necessarily presupposes some metaphysics, and if the metaphysics is bad, or you fail to recognize the effective bounds of that formalism, you can fall into mechanically generated bullshit. Modern predicate logic suffers from known paradoxes and permits nonsensical and vacuous inferences (like those caused by material implication). More subtle effects are expressed in, for example, the problem of bare particulars.

Formalism is a product of prior (semantic) reasoning that isn't formal. And because formalism is syntactic, not only can you still jam your semantic nonsense through it (through incoherent subjects and predicates, for example), but the formalism, stripped of semantics, can itself allow for nonsense. So formalism can actually aid and abet bad reasoning. The danger, of course, is the mistaken notion that "formal = rigorous".

Formalism is also highly impractical and tedious in many circumstances, and it can depart from human reasoning as expressed in the grammar of natural language enough to be practically inscrutable. There is no reason why natural language cannot be clear and well-written. So, I'm afraid you're barking up the wrong tree here.

The problem with LLMs isn't that they're not "formal". It's because they're statistical machines, not reasoning machines, yet many people treat them like magical oracles.