> machine learning models to generate fully correct control software for safety- and mission-critical systems. Using a novel, logic-driven architecture that eliminates hallucinations and errors Pretty curious how this works in practice. It says it's test driven, but then there's no way it will really be complete and correct?