Writing on LLM execution, runtime reliability, and deterministic AI systems.
The model did not get dumber. The failure compounded.
