In all seriousness, if AI models can handle quantization, they can handle some flipped bits from time to time! There are probably some fascinating papers to be written around how to choose which layers in an LLM architecture could benefit more than others from redundant computation in a high-radiation environment.